Interactive guide

68, 95, 99.7: the same three numbers, every normal curve

For any normal distribution, no matter its mean or spread, about 68% of the data sits within one standard deviation of the mean, 95% within two, and 99.7% within three. Those aren't rounded guesses, they're exact areas under the curve, the same three numbers whether σ is 1 or 100. Drag the mean and spread below and watch the bands stay put in standard deviations even as the curve itself moves and stretches.

A property of shape, not of scale

Measure any distance in standard deviations instead of raw units, and a normal curve always looks identical: the same bell, the same proportions, every time. That's exactly why the 68/95/99.7 numbers never change with μ or σ, they're areas under the one true, unchanging shape, just relabeled in whatever units the real data happens to use.

The exact figures are 68.27%, 95.45%, and 99.73%, the area under the standard normal curve between −1 and 1, between −2 and 2, and between −3 and 3. "68-95-99.7" is just the rounded version people actually memorize.

This only holds because the shape is normal. A skewed or bimodal population, like the ones on the Central Limit Theorem page, won't put 68% of its data within one standard deviation at all, until you're looking at the sampling distribution of a large-enough sample mean instead of the raw data itself.

Reading the diagram

  1. 1

    Three nested bands, one curve

    Darkest at the center (±1σ), lightest at the edges (±3σ).

  2. 2

    Drag the value marker

    Its z-score and percentile update live, computed from the real normal CDF, not looked up from a table.

  3. 3

    Generate real samples

    Actual random draws from this exact μ, σ, counted into each band and compared to the theoretical share.

  4. 4

    Watch small samples wobble, large ones settle

    The same Law of Large Numbers idea: more draws, closer to the theoretical 68/95/99.7.

Drag the marker under the curve.

 ±1σ (68.27%)    ±2σ (95.45%)    ±3σ (99.73%)

Marker value
z-score
Percentile

Distribution

50.0
10.0

z = (x−μ)/σ =

Exact vs. rounded rule

BandExactRounded rule

Check it with real samples

BandTheoreticalEmpirical

Draw samples to compare theory against real random data.

Draw only 100 samples and the empirical shares will usually be close but visibly off, 66% here, 71% there, nothing stays exactly 68.27% at that scale. Draw 1,000 and they tighten up considerably. The theoretical 68/95/99.7 numbers describe the exact population; any finite sample is a noisy estimate of it, the same sample-size-versus-precision tradeoff behind the Confidence Intervals page.

The jargons

All of these describe distance from the mean, just measured in different units.

Empirical rule 68-95-99.7 rule

For any normal distribution, roughly 68%, 95%, and 99.7% of the data falls within 1, 2, and 3 standard deviations of the mean.

z-score

(x−μ)/σ: how many standard deviations a value sits from the mean, positive above, negative below.

Percentile

The share of the distribution lying at or below a given value, exactly the normal CDF evaluated there.

Standard normal distribution

The one normal curve with μ=0 and σ=1. Every other normal curve is this same shape, just shifted and rescaled.

Outlier informal

Often flagged past ±2 or ±3 standard deviations, precisely because the empirical rule says so little of the data should be out there.

68/95/99.7 vs. Chebyshev's inequality

The empirical rule needs a normal shape. Chebyshev's inequality gives weaker but universal bounds (at least 75% within 2 SD) for any distribution at all.

A teaching tool: the curve is a real normal probability density function, and every percentage shown, exact or empirical, is computed directly from it or from real pseudo-random draws, not hardcoded.